Bug: ValueError when loading model previously cached but now missing about hub

[singhniraj08]

@sammlapp,

tfhub.dev has been converged with Kaggle Model hub. You can refer this for update. Future improvements will be driven by Kaggle team. Thank you!

[singhniraj08]

@sammlapp,

To avoid the issue of TF-Hub looking in temp directory for cached models, you can customise the download location to home directory by by setting the environment variable TFHUB_CACHE_DIR (recommended) or by passing the command-line flag --tfhub_cache_dir. Users who prefer persistent caching across system reboots can instead set TFHUB_CACHE_DIR to a location in their home directory. When using a persistent location, be aware that there is no automatic cleanup.

I would recommend you to download the model from tfhub.dev with assets, variables and .pb checkpoint file and save it to your home directory and specify the downloaded model folder path in the hub.load() to load the model from local storage instead of looking in temp directory. Example Below

model = hub.load("/Users/singhniraj/Downloads/universal-sentence-encoder-multilingual-large_3/")

Other way would be to instruct the tensorflow_hub library to directly read models from remote storage (GCS) instead of downloading the models locally. This way, no caching directory is needed.

Ref: Caching model downloads from TF Hub. Thank you!

[sammlapp]

Thanks for the workaround. In my case, I am developing a package where the tensorflow hub usage is "under the hood" and not part of the user's experience. I feel it should be possible to avoid this buggy default download behavior without requiring the user to make changes to their TF Hub environment variables.

[singhniraj08]

@sammlapp, Thank you for the feedback. Let us discuss this feature implementation internally and we will update this thread. Thanks!

[sammlapp]

Are there any plans to address this issue? Thanks

[singhniraj08]

Closing this due to inactivity. Please take a look into the answers provided above, feel free to reopen and post your comments(if you still have queries on this). Thank you!